bit-tech.net

How it all shakes out

Introduction

So what does the AMD-ATI merger actually mean? We had all seen the speculation, all read the rumours - but how many people actually expected it to happen? AMD has bought ATI outright, thanks to $2.5 billion of bank money. The four-way power struggle between ATI, AMD, NVIDIA and Intel has now become a three-way, with some interesting dynamics. Why don't we analyse what this means for each of the companies involved? Starting with...

AMD

So why does AMD want to buy ATI? There are a number of answers.

Survival: AMD needs to stay alive. It has not traditionally been a strong company financially. Whilst AMD has always been good at making great gaming processors, the company has not been particularly great at anything else in the past. The one thing it has managed to do in recent times is to create some good server/workstation chips, and those have been very successful in terms of market share, with even Dell adopting certain server configurations.

AMD has had the run of Intel for the past couple of years, as the blue giant made stuff-up after stuff-up. The twilight years of NetBurst were painful for Dell in terms of face, and represented an enormous opportunity for AMD to cannibalise massive amounts of sales. (Note: despite taking a performance beating, Intel was still financially successful by any measurement, since it is a very diverse company).

Now Intel is back with a new architecture that we've seen can trounce AMD in every market. Opteron is being caned by Woodcrest and Athlon 64 is underperforming compared to Conroe. One has to imagine that, now it finally has some decent chips, Intel is going to come back into the game big time.

AMD failed to dent Intel significantly in the time it had, through a lack of strategic planning and bad management. Now Intel is back with superior products, it's in trouble. Unlike Intel, which has a broad portfolio of businesses and interests to generate it profit, AMD pretty much just makes money on microprocessors. This deal not only gives it more weight, it allows it to expand its portfolio of products to be less reliant on just that aspect of its market. The deal makes it harder to kill AMD, and this is a good thing.

"AMD failed to dent Intel significantly in the time it had"

Platform company: This deal makes AMD into a platform company, mimicking Intel. Intel has said for a while now it wants to make platforms, not processors - hence Centrino, Viiv, vPro etc. The ability to create a platform and sell a package of products - CPU, chipset, networking, now graphics - is a powerful thing for system builders. It gives more power to the platform creator and enables them to sell lots of side-products off the back of the success of one main one. AMD, as we mentioned, has always just sold processors, but it can now start on a path to developing wider sales and technology strategies.

Media integration: AMD Live and Intel Viiv are two different ways of approaching the digital home issue. Intel believes in an Intel chip in every computer in the house - in the living room, study, kitchen, wherever. AMD's stated aim with Live is to have a central AMD server machine that interacts with 'thin' set-top boxes around the home.

One of ATI's major money-making divisions is its TV chip division, and this is a good fit which could enable AMD to make strides in the digital home.

GPU on CPU: The biggest aspect of this deal is the integration of GPU functionality onto the CPU. Let's examine this.

No co-processor has ever really survived. Remember maths co-processors on the 386? Integrated in the main processor. Remember motherboard cache? Integrated. Memory controllers? Integrated. Floating point units? Integrated. The CPU, it has been said, is basically a black hole that swallows other technology.

The GPU has done well to survive as a viable co-processor for as long as it has. However, GPUs, in their current form, will not always survive. Whilst there will always be room for ludicrously high-end add-in boards with massive cooling units, we are going to see a shift to graphics being processed by a dedicated graphics core integrated on to the CPU package.

"No co-processor has ever really survived"

How is this possible? Well, a GPU is just another processor. We've seen things go the other way - graphics companies say that their processors can handle physics and other calculations as a 'general purpose GPU'. But what is a general purpose GPU except for a CPU? As functionality aligns and re-aligns, it makes more and more sense to create an integrated chip at the heart of the computer that handles all processing, be it system related, memory related, graphics or physics related. As we gain the ability to put more and more cores on CPUs, expect to see some of those cores dedicated to graphics processing, especially as 3D graphics capability becomes more important (cf Windows Vista).

This isn't a theory; this is actually happening. AMD has even stated this in conversation with analysts and press this morning. Expect to see integrated GPU-CPUs in 2008.

(Note - the exact form of the integration is still up for grabs. The graphics capability could be a general purpose core on the CPU. It could be a GPU core on the CPU. It could be a dual processor motherboard where one of the processors can be a drop-in graphics chip. That is still up for grabs, since not even the engineers know the best answer to that yet).

Those are all very long term goals. In the short term, AMD will acquire a business that is profitable and will nix the increasingly close relationship between Intel and ATI, which up until recently was collaborating on Conroe and future technologies. This will annoy Intel in the short term.